Astrocyte biomarkers GFAP and YKL‐40 mediate early Alzheimer's disease progression

Abstract

INTRODUCTION

We studied how biomarkers of reactive astrogliosis mediate the pathogenic cascade in the earliest Alzheimer's disease (AD) stages.

METHODS

We performed path analysis on data from 384 cognitively unimpaired individuals from the ALzheimer and FAmilies (ALFA)+ study using structural equation modeling to quantify the relationships between biomarkers of reactive astrogliosis and the AD pathological cascade.

RESULTS

Cerebrospinal fluid (CSF) amyloid beta (Aβ)_42/40 was associated with Aβ aggregation on positron emission tomography (PET) and with CSF p‐tau₁₈₁, which was in turn directly associated with CSF neurofilament light (NfL). Plasma glial fibrillary acidic protein (GFAP) mediated the relationship between CSF Aβ_42/40 and Aβ‐PET, and CSF YKL‐40 partly explained the association between Aβ‐PET, p‐tau₁₈₁, and NfL.

DISCUSSION

Our results suggest that reactive astrogliosis, as indicated by different fluid biomarkers, influences the pathogenic cascade during the preclinical stage of AD. While plasma GFAP mediates the early association between soluble and insoluble Aβ, CSF YKL‐40 mediates the latter association between Aβ and downstream Aβ‐induced tau pathology and tau‐induced neuronal injury.

Highlights

• Lower CSF Aβ_42/40 was directly linked to higher plasma GFAP concentrations.

• Plasma GFAP partially explained the relationship between soluble Aβ and insoluble Aβ.

• CSF YKL‐40 mediated Aβ‐induced tau phosphorylation and tau‐induced neuronal injury.

1. BACKGROUND

There is increased recognition that glial cells play an active role in the pathogenesis of Alzheimer's disease (AD). ¹ , ² Astrocytes are important regulators of the brain's inflammatory response to injury and have been shown to become activated in reaction to the deposition of misfolded protein aggregates. ³ Moreover, several studies have demonstrated that reactive astrocytes surround amyloid beta (Aβ) plaques and tau deposits early in AD ⁴ , ⁵ , ⁶ and shown a strong correlation between astrocyte reactivity and increased accumulation of AD pathology. ⁷ , ⁸ , ⁹ These activated astrocytes in turn release pro‐inflammatory molecules such as cytokines and chemokines, which may contribute to neurotoxic effects and exacerbate the progression of AD. ⁴ , ¹⁰ , ¹¹ However, the specific impact of reactive astrogliosis on key pathological events early in the AD continuum remains uncertain. A deeper understanding of how central disease mechanisms are mediated by activated astrocytes may provide us with insight into pathogenic mechanisms underlying AD.

Two robust fluid biomarkers for measuring astrocyte reactivity in vivo are glial fibrillary acidic protein (GFAP) and chitinase‐3‐like protein 1 (YKL‐40), ¹² both of which have consistently been found to be elevated in the dementia phase of AD. ¹³ , ¹⁴ , ¹⁵ Recent work has suggested that changes in astrocytes arise very early in the course of AD, prior to frank neurodegeneration and cognitive impairment, demonstrating an upregulation of GFAP and YKL‐40 levels in Aβ‐positive cognitively unimpaired (CU) individuals. ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² In particular, plasma GFAP, rather than GFAP in cerebrospinal fluid (CSF), has demonstrated superior performance in detecting Aβ‐positive CU individuals. ¹⁷ , ²³ , ²⁴ Moreover, some studies have hypothesized that astrogliosis may even precede the formation of amyloid plaques. ²⁵ , ²⁶ , ²⁷

This Aβ‐induced astroglial response could in turn impact downstream pathological events, including further aggregation of Aβ, tau pathology, neuronal damage, and cognitive decline. ⁴ , ²⁸ , ²⁹ , ³⁰ However, the impact of reactive astrocytes on disease progression has been shown to be very heterogeneous, and reactive astrocytes may respond differently depending on disease stage, specific pathology, biomarker, brain region, and genetic background. ³¹ , ³² , ³³ , ³⁴ Therefore, we aimed to study how two astrocyte biomarkers (plasma GFAP and CSF YKL‐40), probably reflecting different astrocyte phenotypes, mediate the early pathogenic cascade in the preclinical stages of AD. Through structural equation modeling (SEM), we aimed to analyze the relationships among multiple pathological hallmarks of AD, including biomarkers of amyloid pathology (CSF Aβ_42/40 and Aβ‐positron emission tomography [PET]), tau pathology (CSF p‐tau₁₈₁), neuronal damage (CSF NfL), and cognitive performance simultaneously, and test whether and how these relationships are affected by reactive astrocytes. This may help us untangle the complex interplay among pathological changes occurring in the earliest stages of AD.

2. METHODS

2.1. Study participants

Participants were selected from the ALFA+ study, a longitudinal research cohort of CU individuals aged 45 to 74, enriched for a family history of AD or Apolipoprotein E (APOE) ε4 carriership. All participants scored above pre‐established cut‐off values on the following neuropsychological tests: Mini‐Mental State Examination (≥26), Memory Impairment Screen (≥6), Time Orientation Subtest of the Barcelona Test II (≥68), verbal semantic fluency (naming animals ≥12), and a Clinical Dementia Rating (CDR) of 0. A more detailed description of the study protocol can be found in Molinuevo et al. (2016). ³⁵ The study was approved by an independent ethics committee ‘Parc de Salut Mar’, Barcelona, and is registered at Clinicaltrials.gov (Identifier: NCT02485730).

2.2. Fluid biomarker sampling and analysis

CSF samples were obtained by lumbar puncture following a standardized protocol ³⁶ and then collected in 15‐mL polypropylene tubes (Sarstedt catalogue no. 62.554), aliquoted into 0.5‐mL polypropylene tubes (Sarstedt catalogue no. 72.730.005), and frozen at −80°C within 2 h after lumbar puncture. Blood samples were collected using a 20‐ or 21‐g needle gauge into a 10‐mL EDTA tube (BD Hemogard, 10 mL, K2EDTA, catalogue no. 367525). ³⁷ Tubes were gently inverted five to 10 times and centrifuged at 2000 × g for 10 min at 4°C. The supernatant was aliquoted in volumes of 0.5 mL into sterile poly(propylene) tubes (Sarstedt Screw Cap Micro Tube, 0.5 mL, PP, ref. no. 72.730.105) and immediately frozen at −80°C. The samples were processed at room temperature. The time between collection and freezing of both CSF and plasma samples was < 30 min.

CSF p‐tau₁₈₁ was measured using the electrochemiluminescence Elecsys immunoassay on a fully automated cobas e601 module (both Roche Diagnostics International Ltd., Rotkreuz, Switzerland). CSF Aβ₄₀, Aβ₄₂, NfL, and YKL‐40 were measured using the Roche NeuroToolKit immunoassays (Roche Diagnostics International Ltd.) on a cobas e411 or e601 analyzer. Plasma GFAP was quantified on the Simoa HD‐X (Quanterix, Billerica, MA, USA) using the commercial single‐plex assay. All CSF and plasma measurements were performed at the Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, Mölndal, Sweden. A‐T+ individuals, as determined by a CSF Aβ_42/40 ratio > 0.071 (A−) and CSF p‐tau₁₈₁ > 24 pg/mL (T+), were removed from further analysis as they were considered to reflect non‐AD pathological changes. ²²

2.3. Image acquisition and processing

Amyloid PET scans were acquired on a Siemens Biograph mCT scanner, following a cranial computed tomography (CT) scan for attenuation correction. Four frames (4 × 5 min) were collected 90 to 110 min after the injection of 185 MBq [¹⁸F]flutemetamol. ³⁸ An OSEM3D algorithm with eight iterations and 21 subsets was used to reconstruct the images with a point spread function and time‐of‐flight corrections into a 1.02 × 1.02 × 2.03‐mm matrix. The averaged PET images were co‐registered to the corresponding T1‐weighted (T1w) magnetic resonance imaging (MRI) images. Three‐dimensional (3D) high‐resolution T1w turbo field echo (TFE) images (voxel size 1 mm³ isotropic, TR/TE/TI: 6.16/2.33/450 ms, flip angle = 12°) were obtained using a 3T scanner (Ingenia CX, Philips Healthcare, Best, The Netherlands). The T1‐weighted images and co‐registered PET images were then warped to Montreal Neurological Institute (MNI) space with SPM12. The standardized uptake value ratio (SUVR) was calculated in MNI space using the standard target region (https://www.gaain.org/centiloid‐project) with the whole cerebellum as a reference region. We then transformed the SUVR values into the centiloid (CL) scale using a previously calibrated conversion equation. ³⁹ , ⁴⁰ Aβ‐PET was available for a subset of participants (n = 195).

RESEARCH IN CONTEXT

1. Systematic review: We reviewed the literature using PubMed and previously published reviews. Recent publications investigating astrocyte biomarkers and describing the biological mechanisms underlying AD are cited throughout the manuscript.

2. Interpretation: Our results indicate that astrocyte reactivity, as measured by increased plasma GFAP and CSF YKL‐40 concentrations, is associated with the build‐up of Aβ plaques and downstream neurodegenerative events in the earliest stages of the AD continuum.

3. Future directions: Longitudinal studies across the full spectrum of AD are needed to increase our understanding of how the influence of astrogliosis on the progression of AD may change over time and differ by disease stage.

2.4. APOE genotyping

Total DNA was obtained from the blood cellular fraction by proteinase K digestion followed by alcohol precipitation. APOE genotype was obtained from the allelic combination of the rs429358 and rs7412 variants. All participants were classified as APOE ε4 carriers or APOE ε4 non‐carriers.

2.5. Neuropsychological evaluation

In this study, a modified version of the Preclinical Alzheimer Cognitive Composite (PACC) score was used, ⁴¹ which consisted of the Free and Cued Selective Reminding Test (total immediate recall), ⁴² the Logical Memory test of the Wechsler Memory Scale (total delayed recall), ⁴³ the WAIS‐IV Coding subtest, ⁴⁴ and semantic fluency (animals within 1 min). ⁴⁵ All raw test scores were standardized into z‐scores using the mean and standard deviation (SD) from CU A‐T‐ participants as a reference and then averaged into a composite score.

2.6. Statistical analysis

Non‐normally distributed CSF and plasma biomarkers were log₁₀‐transformed. Biomarker values three times outside the interquartile range below Q1 or above Q3 were considered outliers and were removed from further analysis.

To investigate the association between astrocyte phenotypes and pathological hallmarks of AD, we applied linear regression analysis using astrocyte biomarkers as predictors and individual core AD and neurodegeneration biomarkers as outcomes while adjusting for age, sex, and APOE ε4 carriership. Additionally, raw associations between all model parameters were examined by performing a cross‐correlation using Pearson's r.

To model a potential mediating role of reactive astrogliosis (ie, elevations in plasma GFAP and CSF YKL‐40) on the association between biomarkers in the AD cascade (ie, changes in amyloid, tau, and neuronal injury biomarkers), we built a path model. Our model was hypothesis‐based with reference to the current literature and followed the amyloid neuropathological cascade pathway. ²⁹ , ⁴⁶ , ⁴⁷ , ⁴⁸ To quantify complex multivariate relationships in our AD cascade model simultaneously, we used SEM. In contrast to standard regression modeling in which all variable coefficients are calculated separately, each structural equation coefficient is computed while considering the direct and indirect effects between all biomarkers. Therefore, SEM makes it possible to test more complicated mediation models in a single analysis. Our model is hierarchical in nature and structured in seven levels: (I) covariates age, sex, and APOE ε4 carriership; (II) CSF Aβ_42/40 ratio as an initial pathological trigger; (III) astrocytic biomarkers, which may serve as potential mediators along all associations in this cascade; (IV) Aβ‐PET global CL level; (V) CSF p‐tau₁₈₁ level; (VI) NfL concentration in CSF; and finally (VII) cognitive performance as measured by the PACC. In SEM, a variable can appear as a predictor in several equations as well as the outcome in others. The direct effect of a predictor variable on a higher‐level outcome can be interpreted as the net effect of a predictor when adjusting for the other predictors (ie, variables of preceding levels) in the equation, and is visualized by an arrow in the path model. The indirect effect is the effect mediated by the reactive astrocyte variables.

All variables in the model were standardized (z‐score), so that 1 SD change in the independent variable predicts 1 SD change in the dependent variable (while holding the other variables in the model constant) and to allow for direct comparisons of beta estimates. The model parameters (effects) were computed by maximum likelihood estimation. Effects were considered significant at p < 0.05 false discovery rate (FDR) corrected. The 95% confidence interval (CI) of the parameters was estimated using Monte Carlo bootstrapping (1000 iterations). Model fit was evaluated by a comparative fit index ≥ 0.90, which indicates that the model accounts for most of the variance in the data and by a root mean square error of approximation statistic < 0.05, indicating low residual values not accounted for by the model. ⁴⁹ All statistical analyses were carried out using R version 4.2.2, with the lavaan package for SEM analysis. ⁵⁰

3. RESULTS

The sample characteristics are summarized in Table 1. There were a total of 384 CU participants, 61% of whom were female, and 54% were APOE ε4 carriers; the mean age was 61 years old. The average Mini‐Mental State Examination (MMSE) score was 29.2 (range 27.0 to 30.0), and centiloid values ranged from −21.15 to 63.10, with a mean value of 4.39.

TABLE 1.

Participant characteristics.

	Total (N = 384)
Age (y)	61.1 [49.6 to 73.4]
Sex, F	235 (61.2%)
Education (y)	13.5 (3.5)
MMSE	29.2 [27.0 to30.0]
PACC (z‐score)	0.00 [−2.01 to 1.40]
APOE ε4 carrier	208 (54.2%)
Aβ‐PET (CL)*	4.39 [−21.15 to 63.10]
Plasma GFAP (pg/mL)	94.8 (44.4)
CSF Aβ42/40	0.07 (0.02)
CSF YKL‐40 (pg/mL)	145.3 (51.7)
CSF p‐tau181 (pg/mL)	307.7 (141.6)
CSF NfL (pg/mL)	80.7 (25.7)

Note: Data is presented as mean (SD), mean [range], or N (%); * n = 195.

Abbreviations: Aβ, β‐amyloid; APOE, apolipoprotein E; CL, centiloid; CSF, cerebrospinal fluid; F, female; GFAP, glial fibrillary acidic protein; MMSE, Mini‐Mental State Examination; NfL, neurofilament light; PACC, Preclinical Alzheimer Cognitive Composite; PET, positron emission tomography; p‐tau, phosphorylated tau; y, years; YKL‐40, chitinase‐3‐like protein 1.

Linear regression analysis of the relationship of astrocyte biomarkers with core AD and neurodegeneration biomarkers (Figure 1; Table 2) revealed that higher CSF YKL‐40 was associated with higher Aβ centiloid values, higher CSF p‐tau₁₈₁, and higher CSF NfL. Higher CSF YK‐L40 was also weakly associated with a lower Aβ_42/40 ratio and worse cognitive performance. Higher plasma GFAP was associated with lower CSF Aβ_42/40, higher Aβ‐PET load, and higher CSF p‐tau₁₈₁ and showed a weaker association with higher CSF NfL. In addition, plasma GFAP showed no significant association with cognitive performance or with CSF YKL‐40 after adjusting for age, sex, and APOE ε4 carriership. Additional analysis using a subset of participants (n = 195) with all biomarker measurements available showed similar results (Figure S1).

FIGURE 1.

Scatterplots showing the relation between astrocyte biomarkers (ie, YKL‐40 and GFAP) with biomarkers of the AD cascade (ie, Aβ_42/40, Aβ‐PET, p‐tau₁₈₁, NfL, PACC). All models included age, sex, and APOE ε4 allele status. Abbreviations: Aβ, β‐amyloid; CL, centiloid; CSF, cerebrospinal fluid; GFAP, glial fibrillary acidic protein; NfL, neurofilament light; PACC, Preclinical Alzheimer Cognitive Composite; PET, positron emission tomography; p‐tau, phosphorylated tau; YKL‐40, chitinase‐3‐like protein 1.

TABLE 2.

Linear regression analysis: Individual associations between CSF YKL‐40 and plasma GFAP with AD cascade biomarkers.

	CSF YKL‐40	Plasma GFAP
CSF Aβ42/40	−0.00005 (0.00002)*	−0.00016 (0.00002)****
Aβ‐PET (CL)	0.09 (0.02)***	0.15 (0.02)****
CSF p‐tau181	1.29 (0.14)****	0.99 (0.17)****
CSF NfL	0.22 (0.02)****	0.09 (0.03)**
PACC	−0.0015 (0.0007)*	0.0008 (0.0008)

Note: Data are represented as beta (standard error); betas are unstandardized. All models included age, sex, APOE ε4 allele status.

Abbreviations: Aβ, β‐amyloid; CL, centiloid; CSF, cerebrospinal fluid; GFAP, glial fibrillary filament protein; NFL, neurofilament light; PACC, Preclinical Alzheimer Cognitive Composite; p‐tau, phosphorylated tau; YKL‐40, Chitinase 3‐like 1.

* p < 0.05; ** p < 0.01; *** p < 0.001; **** p < 0.0001.

3.1. Structural equation model

The results of our path model are shown in Figure 2, which displays all significant direct associations between the biomarkers. All possible (ie, significant and non‐significant) associations and their corresponding estimates with 95% CI are shown in Table 3. We observed that older age (β = −0.24; 95% CI = −0.32 to −0.15) and particularly APOE ε4 carriership (β = −0.74; 95% CI = −0.92 to −0.56) showed a direct significant association with lower CSF Aβ_42/40 ratio. Moreover, older age (β = 0.36; 95% CI = 0.27 to 0.45) and early Aβ pathology, as indicated by lower CSF Aβ_42/40 ratio (β = −0.33; 95% CI = −0.43 to −0.23), were directly related to increased plasma GFAP. Older age (β = 0.37; 95% CI = 0.27 to 0.47), higher Aβ load on PET (β = 0.21; 95% CI = 0.10 to 0.33), higher CSF p‐tau₁₈₁ (β = 0.31; 95% CI = 0.22 to 0.39), and increased CSF NfL (β = 0.18; 95% CI = 0.07 to 0.27) were directly associated with higher CSF YKL‐40. There was a strong direct effect of lower CSF Aβ_42/40 ratio (β = −0.54; 95% CI = −0.68 to −0.40) on elevated Aβ‐PET and, to a lesser extent, from plasma GFAP (β = 0.17; 95% CI = 0.03 to 0.31) and from CSF YKL‐40 (β = 0.19; 95% CI = 0.06 to 0.33) on Aβ‐PET. Furthermore, a direct association was observed between CSF p‐tau₁₈₁ and CSF Aβ_42/40 (β = −0.26; 95% CI = −0.40 to −0.13), Aβ‐PET load (β = 0.31; 95% CI = 0.13 to 0.48), and CSF YKL‐40 (β = 0.40; 95% CI = 0.30 to 0.50). In addition, older age (β = 0.20; 95% CI = 0.11 to 0.30), male sex (β = 0.57; 95% CI = 0.41 to 0.74), higher CSF p‐tau₁₈₁ (β = 0.18; 95% CI = 0.06 to 0.31), and higher CSF YKL‐40 (β = 0.31; 95% CI = 0.20 to 0.41) were directly associated with increased CSF NfL. Finally, older age (β = −0.19; 95% CI = −0.28 to −0.11) was directly associated with worse cognitive performance, as measured by the PACC score. Note that in SEM, all tested associations are corrected for parameters of preceding levels. A descriptive cross‐correlation matrix across all model parameters separately is provided in Supplementary Figure 2.

FIGURE 2.

Path analysis showing the impact of CSF YKL‐40 and plasma GFAP on Amyloid‐Tau‐Neurodegeneration and cognition. Cognition was measured by the Preclinical Alzheimer Cognitive Composite (PACC) score. Arrows show the direct effects of significant associations at p < .05 FDR‐corrected between all biomarker relationships (z‐score) from the structural equation model. The beta estimates represent the unique contribution of a specific variable to the change in a dependent variable after controlling for the effects of all variables of preceding levels in the model.

TABLE 3.

Structural equation model coefficients of Figure 2 displaying all direct and indirect effects of the path model.

Biomarker	β (95% CI)	p value
Cognition (PACC)
NfL	0.014 (−0.074 to 0.114)	0.974
p‐tau181	−0.015 (−0.118 to 0.091)	0.974
Aβ‐PET	0.027 (−0.120 to 0.155)	0.974
YKL‐40	−0.078 (−0.173 to 0.018)	0.358
GFAP	0.031 (−0.045 to 0.113)	0.974
Aβ42/40	0.025 (−0.080 to 0.123)	0.974
Sex, male	0.100 (−0.049 to 0.243)	0.511
Age	−0.192 (−0.280 to −0.109)	<0.001*
APOE ε4 carrier	0.098 (−0.04 to 0.235)	0.471
NfL
p‐tau181	0.182 (0.057 to 0.305)	0.015*
Mediated by YKL‐40	0.094 (0.055 to 0.138)	<0.001*
Aβ‐PET	0.076 (−0.065 to 0.225)	0.505
YKL‐40	0.306 (0.196 to 0.406)	<0.001*
GFAP	0.072 (−0.009 to 0.154)	0.225
Aβ42/40	0.027 (−0.090 to 0.133)	0.796
Sex, male	0.570 (0.410 to 0.739)	<0.001*
Age	0.202 (0.107 to 0.304)	<0.001*
APOE ε4 carrier	0.009 (−0.150 to 0.176)	0.939
p‐tau181
Aβ‐PET	0.305 (0.133 to 0.481)	0.002*
Mediated by YKL‐40	0.082 (0.041 to 0.139)	0.002*
YKL‐40	0.398 (0.302 to 0.497)	<0.001*
GFAP	0.059 (−0.036 to 0.150)	0.294
Aβ42/40	−0.261 (−0.404 to −0.131)	<0.001*
Sex, male	0.055 (−0.117 to 0.208)	0.564
Age	−0.061 (−0.154 to 0.030)	0.276
APOE ε4 carrier	0.002 (−0.171 to 0.170)	0.999
Aβ‐PET
YKL‐40	0.194 (0.059 to 0.328)	0.008*
GFAP	0.170 (0.034 to 0.312)	0.030*
Aβ42/40	−0.541 (−0.677 to −0.402)	<0.001*
Mediated by GFAP	−0.055 (−0.106 to −0.009)	0.043*
Sex, male	−0.061 (−0.296 to 0.171)	0.635
Age	0.010 (−0.119 to 0.129)	0.896
APOE ε4 carrier	−0.104 (−0.328 to 0.130)	0.430
YKL‐40
NfL	0.177 (0.067 to 0.274)	0.004*
p‐tau181	0.309 (0.220 to 0.392)	<0.001*
Aβ‐PET	0.206 (0.103 to 0.325)	0.001*
Aβ42/40	−0.085 (−0.196 to 0.033)	0.193
Sex, male	−0.188 (−0.396 to 0.036)	0.113
Age	0.370 (0.269 to 0.466)	<0.001*
APOE ε4 carrier	−0.112 (−0.306 to 0.106)	0.368
GFAP
NfL	0.008 (−0.03 to 0.115)	0.965
p‐tau181	0.014 (−0.065 to 0.089)	0.823
Aβ‐PET	0.107 (−0.001 to 0.225)	0.124
Aβ42/40	−0.327 (−0.429 to −0.228)	<0.001*
Sex, male	−0.084 (−0.266 to 0.090)	0.430
Age	0.361 (0.274 to 0.446)	<0.001*
APOE ε4 carrier	−0.167 (−0.367 to 0.030)	0.142
Aβ42/40
Sex, male	0.091 (−0.102 to 0.271)	0.430
Age	−0.235 (−0.315 to −0.152)	<0.001*
APOE ε4 carrier	−0.743 (−0.922 to −0.562)	<0.001*

Note. Structural equation model showing standardized coefficients with bootstrapped 95% confidence intervals. Model shown in Figure 2. N = 384 (Aβ‐PET n = 195). P values adjusted for multiple comparisons using false discovery rate were considered significant at p < 0.05.

Abbreviations: Aβ, β‐amyloid; APOE, apolipoprotein E; CI, confidence interval; GFAP, glial fibrillary acidic protein; NfL, neurofilament light; PACC, Preclinical Alzheimer Cognitive Composite; PET, positron emission tomography; p‐tau, phosphorylated tau; YKL‐40, chitinase‐3‐like protein 1.

3.2. Mediation effects

We observed that part of the relationship between CSF Aβ_42/40 and Aβ‐PET load could be explained by plasma GFAP (Figure 2; Proportion mediated 5%; β = −0.06; 95% CI = −0.11 to −0.01). Furthermore, CSF YKL‐40 mediated the association of Aβ‐PET with CSF p‐tau₁₈₁ (Proportion mediated 8%; β = 0.08; 95% CI = 0.04 to 0.14), as well as the association of CSF p‐tau₁₈₁ with CSF NfL (Proportion mediated 9%; β = 0.09; 95% CI = 0.06 to 0.14). Since no direct association between CSF Aβ_42/40 and CSF YKL‐40 was observed, no CSF Aβ_42/40 induced mediation by CSF YKL‐40 on other biomarkers in the cascade was present. Similarly, plasma GFAP showed no direct relationship with CSF p‐tau₁₈₁, and thus no mediation by GFAP on the effect of CSF Aβ_42/40 on CSF p‐tau₁₈₁ was established.

4. DISCUSSION

Our structural equation model revealed that plasma GFAP and CSF YKL‐40 are important mediators of key events in the AD cascade and strongly contribute to the progression of AD at an early stage of the disease. We observed that the earliest abnormalities in CSF Aβ_42/40 triggered an upregulation of GFAP in the blood. The association between CSF Aβ_42/40 and Aβ‐PET was partially explained by this increase in plasma GFAP. This suggests that astroglia, particularly those that release GFAP, may have a role in the early balance between soluble and insoluble Aβ aggregates. Furthermore, we observed that the release of YKL‐40 into the CSF occurred slightly later in the pathological cascade and was linked with Aβ‐induced tau phosphorylation and tau‐induced axonal damage. These results further support the evidence that reactive astrogliosis is an early event in AD and a significant component of the pathological cascade driving neurodegeneration.

Our results are in agreement with increasing evidence that suggests that astrocyte changes occur very early in the course of AD. Furthermore, they indicate that distinct stages of the early pathological cascade in preclinical AD are associated with GFAP and YKL‐40 upregulation, which show differential responses to pathological stimuli. This is in line with previous studies showing distinct astrocyte biomarker signatures in response to Aβ and tau pathology, ³¹ as well as across disease progression. ⁵¹

A strong relationship between Aβ aggregation and GFAP, in blood as well as in CSF, has been frequently demonstrated. ²³ , ⁵¹ , ⁵² We add to these findings by showing for the first time, to our knowledge, that changes in plasma GFAP are partly responsible for the relationship between soluble Aβ and increased fibrillar Aβ deposition on PET. Previous studies demonstrated that a rise in Aβ oligomers was highly associated with astrocyte reactivity. ²⁵ , ²⁶ , ⁵³ Furthermore, it has been shown that astrocytes are involved in the clearance and degradation of Aβ, ⁵⁴ , ⁵⁵ with evidence suggesting that astrocytes can internalize Aβ oligomers and protofibrils but may eventually become overwhelmed and fail to clear Aβ effectively. Moreover, when astrocytes break down, they release the Aβ they have accumulated, actively contributing to the overall accumulation of Aβ plaques. ⁵⁶ , ⁵⁷ , ⁵⁸ Taken together, this implies that astrocytes’ protective mechanisms become impaired in AD and rather gain a neurotoxic function instead. ¹¹ , ²⁵

Recently, Bellaver et al. (2023) ⁹ reported that reactive astrocytes were a prerequisite for the phosphorylation of tau in Aβ‐positive CU patients. Although we used a different study population and approach, that is, binary versus continuous astrocyte measurements, we demonstrated compatible results and provided more granularity on the underlying mechanism. Our results suggest that increased astrocyte reactivity may influence the balance between soluble and insoluble Aβ and that this formation of Aβ plaques in turn triggers tau phosphorylation in preclinical AD.

CSF concentrations of YKL‐40 have been thought to mainly reflect a response to tau pathology rather than Aβ, and strong correlations with tau pathophysiology have also been found in preclinical stages. ¹⁹ , ²² , ³¹ , ⁵⁹ , ⁶⁰ A growing body of studies have demonstrated a positive association of CSF YKL‐40 levels with markers of neuronal injury, including cortical atrophy, CSF t‐tau, and NfL in early stages of AD. ¹⁶ , ²¹ , ²⁸ , ⁶¹ These findings suggest that CSF YKL‐40 is particularly related to tau pathology and neuronal injury and support the hypothesis that reactive astrocytes actively contribute to the disruption of neuronal functioning. ⁴ , ⁶² , ⁶³ , ⁶⁴ Moreover, some studies, including our CU cohort, have suggested YKL‐40 may be involved in a non‐amyloid‐related pathway, demonstrating elevated CSF levels of YKL‐40 in A+T+ and A−T+ individuals, but not in A+T− individuals. ¹⁶ , ²² , ³¹ However, in the current study we observed that fibrillar deposits of Aβ could trigger the expression of YKL‐40, and associations between YKL‐40 levels and Aβ‐PET were also previously reported. ⁶⁵ , ⁶⁶ YKL‐40 immunoreactivity was demonstrated to be independent of tau in a recent post mortem study, ⁶⁰ suggesting that the astrocytic responses by GFAP and YKL‐40 might be more complex than an amyloid‐tau dichotomy. ⁶⁷

In line with previous reports, we observed no direct association between APOE ε4 carriership or sex with astrocyte biomarkers. ¹⁵ , ¹⁸ , ²¹ However, as reported in previous studies, higher CSF NfL concentrations were observed in men. ⁶⁸ , ⁶⁹ Furthermore, aging contributed significantly to both YKL‐40 and GFAP concentrations, as reported previously. ¹⁶ , ¹⁷ , ¹⁸ , ²¹ Finally, we found that age was the only factor that was directly related to cognitive performance, while several previous studies did found an association between GFAP and YKL‐40, both in blood and CSF, and cognition. ³⁰ , ⁷⁰ , ⁷¹ The cross‐sectional nature of this study, the early stage on the AD continuum as reflected by the very low mean CL values, and the limited variance in test scores of cognitively normal participants may explain this finding. Additionally, it is important to note that in the current structural equation model, the association between astrocytosis and cognition is adjusted for the effect of Aβ, tau pathology, and NfL.

Our findings, together with previous evidence, indicate that astrogliosis contributes to the pathogenesis of AD through multiple routes/pathways, which can be observed at early asymptomatic stages of AD. Astroglial response may occur in different stages of preclinical AD: the aggregation of Aβ, the formation of tau tangles, and neuronal damage. It can be speculated that once astrocyte activation is induced, there is a release of pro‐inflammatory molecules and neuronal dysfunction, which in turn reactivates astrocytes. Eventually, this astroglial response may contribute to neurodegenerative changes independently of Aβ plaque pathology. ⁴ , ⁷² , ⁷³ Taken together, these findings suggest that interventions targeting astrocyte dysfunction involved in Aβ clearing in early preclinical stages may ultimately prevent or delay the onset of AD dementia.

There are some limitations to this study. First, the data were collected cross‐sectionally from CU individuals only. This prevented us from being able to make any claims on the causality of the pathological events. Longitudinal studies across the entire AD continuum are needed to provide information on how the interaction between reactive astrocytes and AD pathology markers may change over time, which is important considering the indications that the role of reactive astrocytes in disease progression likely varies across disease stages. ⁵¹ , ⁷² Second, our path model provides a simplified view of the AD pathological cascade and is by no means a saturated depiction of its complex pathophysiology. To illustrate, multiple lines of evidence suggest an important interaction between microglia and astrocytes that likely acts in a coordinated manner to promote the progression of AD. ⁷⁴ The strengths of this study include the well‐characterized cohort and the fact that the SEM approach allowed us to examine all direct and indirect effects on each variable in a single model, rather than studying all relationships separately. This is essential in a multifactorial disease such as AD, in which a complex cascade of connected events ultimately contributes to progression.

In conclusion, we provide evidence that the astrocytosis biomarkers plasma GFAP and CSF YKL‐40 increase very early in the AD continuum and mediate several associations between key pathogenic events that occur during this disease stage. These results substantiate the notion that reactive astrocytes in reaction to AD pathology are active players in promoting downstream neurodegenerative events.

CONFLICT OF INTEREST STATEMENT

W.P., M.S., A.B.S., C.M., K.F., A.G.E., N.J.A. have nothing to disclose. J.L.M. is currently a full‑time employee of H. Lundbeck A/S and previously served as a consultant or on advisory boards for the following for‑profit companies or has given lectures in symposia sponsored by the following for‑profit companies: Roche Diagnostics, Genentech, Novartis, Lundbeck, Oryzon, Biogen, Lilly, Janssen, Green Valley, MSD, Eisai, Alector, BioCross, GE Healthcare, and ProMIS Neurosciences. G.K. is a full‑time employee of Roche Diagnostics GmbH. M.C. is a full‑time employee of Roche Diagnostics International Ltd. and an owner of shares in Roche. H.Z. has served at scientific advisory boards and/or as a consultant for Abbvie, Acumen, Alector, Alzinova, ALZPath, Annexon, Apellis, Artery Therapeutics, AZTherapies, CogRx, Denali, Eisai, Nervgen, Novo Nordisk, Optoceutics, Passage Bio, Pinteon Therapeutics, Prothena, Red Abbey Labs, reMYND, Roche, Samumed, Siemens Healthineers, Triplet Therapeutics, and Wave, has given lectures at symposia sponsored by Cellectricon, Fujirebio, Alzecure, Biogen, and Roche, and is a cofounder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program. K.B. has served as a consultant, at advisory boards, or at data monitoring committees for Abcam, Axon, Biogen, JOMDD/Shimadzu. Julius Clinical, Lilly, MagQu, Novartis, Prothena, Roche Diagnostics, and Siemens Healthineers, and is a cofounder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program. M.S.C. has served as a consultant and at advisory boards for Roche Diagnostics International Ltd. and has given lectures at symposia sponsored by Roche Diagnostics, S.L.U. and Roche Farma, S.A. J.D.G. receives research funding from Roche Diagnostics and GE Healthcare and has given lectures at symposia sponsored by Biogen and Philips. G.S.‐B. has served as a consultant for Roche Farma, S.A. O.G.R. receives research funding from F. Hoffmann‐La Roche Ltd. and has given lectures in symposia sponsored by Roche Diagnostics, S.L.U. Author disclosures are available in the supporting information.

CONSENT STATEMENT

All participants provided written informed consent.

Supporting information

Supporting Information

Supporting Information

Pelkmans W, Shekari M, Brugulat‐Serrat A, et al. Astrocyte biomarkers GFAP and YKL‐40 mediate early Alzheimer's disease progression. Alzheimer's Dement. 2024;20:483–493. 10.1002/alz.13450